I looked thru the doctroral thesis of Lytro's CEO.

Lots of math. (way beyond me) but the basics are that you take a many megapixel sensor, the same as used in many cameras, and install an array of lenslets just in front of the sensor. This has the effect of creating an array of many small images focused on the sensor.

Each sub image covers a dozen or so pixels. All are slightly different from each other. The full result can then be post processed to generate an image. Adjusting various parameters in the math allows reconstruction of the image at various focal distances.

One issue in building the camera is, there is a relationship between the image sharpness that can be achieved vs. the ability to resolve the focal depth, is tied to the distance between the lens array and the sensor. The thesis has examples.

Another interesting bit is, since standard sensors are being used, there is no reason that video can't be captured. I'm guessing that the amount of post processing needed for video is extreme.

More megapixels would seem to be the answer to things like small images and sharper images with the ability to selectively focus thru greater depth.

If photography, from the very beginning, had been able to resolve everything in focus, a lot of what we call the photographic art, might not exist.